Building healthy Lagrangian theories with machine learning

Abstract

The existence or not of pathologies in the context of Lagrangian theory is studied with the aid of Machine Learning algorithms. Using an example in the framework of classical mechanics, we make a proof of concept, that the construction of new physical theories using machine learning is possible. Specifically, we utilize a fully-connected, feed-forward neural network architecture, aiming to discriminate between “healthy” and “nonhealthy” Lagrangians, without explicitly extracting the relevant equations of motion. The network, after training, is used as a fitness function in the concept of a genetic algorithm and new healthy Lagrangians are constructed. These new Lagrangians are different from the Lagrangians contained in the initial data set. Hence, searching for Lagrangians possessing a number of pre-defined properties is significantly simplified within our approach. The framework employed in this work can be used to explore more complex physical theories, such as generalizations of General Relativity in gravitational physics, or constructions in solid state physics, in which the standard procedure can be laborious.